1. Field of the Invention
The present invention relates generally to processing, model extraction, model refinement and graphical rendering of digital data representing geologic or other types of volumes and, more specifically, to rendering and analysis of irregularly shaped volumes within voxel-based volumetric data.
2. Description of the Related Art
Geologists, geophysicists and others analyze seismic data for purposes such as detecting the presence and change over time of hydrocarbon reservoirs or other subsurface features. Seismic data can be gathered by, for example, creating explosions or otherwise releasing energy into the ground, and detecting and digitizing the reflected seismic energy using an array of geophones or similar sensors. The processed volumetric seismic data represent a three-dimensional (3D) subsurface region, typically expressed in time or depth units. Other examples of ways in which such volumetric data can be gathered and used include gravitational and magnetic measurement. The data can comprise any of a large number of attributes that practitioners in the art have identified as being usable or derivable from reflected seismic energy and field measurements, the most common of which perhaps being amplitude of the reflected signals.
Collected 3D or volume datasets can be interpreted, stored and otherwise manipulated in any of a number of formats. An increasingly common format is that in which each data element itself represents a volume. Such a data element is known as a voxel (for “volume element”) or cell. If, for example, amplitude is the attribute that characterizes the collected data, the attribute samples are represented by voxels, each characterized by an amplitude. In other words, the dataset is made up of a multiplicity of voxels, each characterized by an amplitude. A property of such cells may be characterized as a data-collection attribute, from which the volume dataset was constructed, such that the attribute data includes at least one value of the data-collection attribute of the identified cell. A seismic volume dataset commonly comprises millions or even billions of voxels and can require terabytes of data storage. Voxel formats are commonly used not only in seismic data analysis but also in medical imaging and other imaging disciplines.
The analysis of volumetric data typically involves rendering, interpreting, and refining stages to produce a sub-surface model or to render a specific 3D view of the sub-surface region. Most commercially available 3D graphics engines (e.g., graphics accelerator cards for workstation computers) do not have voxel primitives; rather, they can interpret only points, lines, and polygons such as triangles, because they are intended for rendering surface-based representations of 3D objects, i.e., hollow shells, not objects comprising voxels. Although some voxel-based graphics accelerators exist, they cannot efficiently and cost-effectively combine 2D and 3D primitives, which are needed to implement certain display features, such as spherical bill-boarding and animation.
Thus, the known methods for rendering voxel data, such as raycasting and splatting, merely produce snapshot 3D images from some predetermined viewing perspective. In raycasting, rays are projected from the viewer's origin, the user's point of view, or through a projection plane and extended until they intersect a data point or a series of data points. These point(s) along and around the rays are processed and the resulting image drawn on the display or projection plane. In splatting, the contribution to the final image of each voxel around a predetermined neighborhood is computed. Voxels are conceptually or virtually “thrown” onto the image plane such that each voxel in the object space leaves a footprint in the image space. Whether used to render surface-based representations of three-dimensional objects or actual, i.e., voxel-based, three-dimensional objects, these methods can render only flat voxel approximation, i.e., circular or square in shape with no defined voxel neighborhood, and are strictly view-dependent. Also note that all of the known voxel data rendering methods are computationally software or hardware resource intensive and require expensive and specialized hardware that possesses various performance and viewing limitations. In addition, the performance of many of these algorithms is affected by how they traverse memory. When memory is traversed in different directions, the algorithm performance can vary depending upon hardware data access efficiency. The bigger the working data volume the lesser the ability to efficiently use hardware caches.
Furthermore, regardless of which of the various voxel rendering methods is used, the entire volume dataset must be retained in (hardware) memory and eventually rendered by the 3D graphics hardware. To render a data volume, the entire dataset must be retrieved/swapped from data storage (i.e., random access memory, disk memory, etc.) and rendered using computationally intensive algorithms and sent to normally resource-limited 3D graphics hardware. As a result of these limitations, rendering can take a substantial amount of time, depending upon the hardware used. Rendering quality, interactive high-end display options and interactive display response are of considerable concern to persons such as geologists and geophysicists who may wish to view a model from many different directions or viewpoints, compare different models to each other, and otherwise manipulate them in their analysis of subsurface features relating to oil and gas reservoirs.
It is commonly desirable to identify, isolate, and focus upon specific features and/or anomaly regions within a volume, such as those relating to potential oil and gas reservoirs in seismic volumes or those relating to organs, bones and tumors in medical volumes. Such features can be referred to as irregular volumes because they do not have a regular or predictable shape. Raycasting, point splatting and even newer pure voxel rendering schemes cannot readily separate or otherwise work with such irregular volumes separately from the whole working data volume, because in most cases (i.e., when using hardware 3D or 2D texturing) the whole working data volume is eventually sent and completely rendered by the 3D graphics hardware. The image created by the 3D hardware may graphically and partially isolate certain features on the screen, but that is little more than an ephemeral or snapshot view of the data volume from a single specific direction. That is, even though a user may see a feature in the image, the feature is not represented separately within the computer logic apart from the neighboring data. It may be, however, that a user is interested in only a specific irregular volume or group of irregular volumes and not in the volume as a whole, or may be interested in performing some specific attribute mapping operation or view upon only a specific irregular volume or set of irregular volumes. For example, one may wish to consider a series of images, each representing the same irregular volume or group of irregular volumes at a different time, to assess how the irregular volume(s) may have changed over time as, for example, oil or gas is depleted. Conventional rendering systems require one to render each volume dataset in its entirety to see how the irregular volume of interest represented within it appeared at the time the dataset was collected.
Practitioners in the art have attempted to overcome the inability to work with individual irregular volumes by laboriously identifying them using the snapshot views, and extracting each of them from volume datasets. This process can be very painful and tedious if there is a lot of noise in the data or if the objects are numerous and small in size. For example, algorithms have been suggested by which one, using various snapshot views from various angles, visually identifies a seed point or seed voxel belonging to an irregular volume and then uses separate custom-written software to extract all of the mutually neighboring voxels that are presumed to belong to the same irregular volume. Nevertheless, it becomes inefficient, laborious and time-consuming to use such a process and workflow to identify a large number of irregular volumes. The problem is greatly magnified if one is interested in identifying all of the irregular volumes in a very large dataset. Moreover, no practical and efficient means for further manipulating or analyzing irregular volumes extracted in this manner have been suggested.
It would be desirable to provide an efficient system for identifying, rendering and otherwise working with individual irregular volumes represented within one or more volume datasets. The present invention addresses these problems and deficiencies and others in the manner described below.